DRAFT, PLEASE DO NOT CITE

POLITICAL CENTRALIZATION AND LOCAL DEMOCRACY: EVIDENCE FROM LARGE-SCALE MUNICIPAL REFORM§ David Dreyer Lassen Department of Economics University of Copenhagen 1455 Copenhagen K, Denmark [email protected] Søren Serritzlew Department of Political Science University of Aarhus 8000 Aarhus C, Denmark [email protected]

November 14, 2008

Abstract: The question of how jurisdiction size affects democracy is classical and contested. Due to problems of endogeneity and to sorting effects, cross sectional studies yield ambiguous results. In this study we identify a natural experiment of jurisdictional consolidation: A large-scale exogenous reform of municipal government in Denmark, affecting some, but not all municipalities. Based on repeated survey data collected before and after the reform, we find, using various differences-in-differences and matching estimators, that jurisdiction size has a causal detrimental effect on citizens’ sense of internal political efficacy.

Keywords: political participation, local democracy, natural experiment, matching, differences-in-differences, municipal politics, jurisdictional reform, political efficacy

§

We thank Poul Erik Mouritzen for access to his data, KREVI and our respective Departments for

funding, KREVI for facilitating the collection of data, and Søren Leth-Pedersen and participants at MPSA 2008 for comments and suggestions. We also thank Jens Blom-Hansen, Jørgen Elklit, Ulrik Kjær, Jens Peter Frølund Thomsen and Lise Togeby for comments to the questionnaire.

0

I. Introduction How government is structured horizontally and vertically matters for economic and political outcomes. Since Aristotle, political philosophers, political scientists and economists have debated the optimal size of a polity. Most recently, decentralization as a solution to problems of economic and political development has been given substantial support by the World Bank (e.g. World Bank, 2000). Across the world, local, regional and federal governments are constantly (re-)assessing the optimal balance between (more) centralization and (more) decentralization. Designing the optimal structure or architecture of government resolves around a key trade-off:1 Larger, and thereby fewer, jurisdictions allow governments to take advantage of economies of scale in local public goods production. If such production involves fixed costs, such as fixed administrative or bureaucratic costs, having more jurisdictions entails duplicating such fixed costs, which is inoptimal from a societal point of view. At the same time, more jurisdictions result in greater interjurisdictional spill-overs in the form of externalities, arising when governments do not take into account the effects of their decisions on other entities of government. On the other hand, having a larger number of – on average smaller – subnational political units also confers several advantages. The most celebrated advantage of decentralization is that it can improve service delivery of local public goods.2 It does

1

Treisman (2007, pp. 11-15) provides a detailed discussion of these core arguments for or against

decentralization. The rest of his book expands on these arguments and provides an in-depth, critical study of the economics and politics of decentralization. Oates (1972) provides the first full specification of the problem of the optimal size of jurisdictions, trading off economies of scale with heterogeneity of preferences. 2

The existing empirical evidence is mixed. Faguet (2004) provides an overview and an empirical

illustration from Bolivia.

1

so by a combination of two factors. First, having several municipalities improves allocative efficiency or preference-matching by allowing citizens to sort into different jurisdictions based (in part or in whole) on their preferences over local public goods; this is the classical Tiebout (1956)-argument.3 Second, having a more decentralized structure is also thought to improve productive efficiency. This can happen through a number of channels: Smaller jurisdictions can increase electoral control and accountability (e.g. Seabright, 1996; Hindricks and Lockwood, 2005). Having several jurisdictions can, by increasing the competition for tax payers and firms, lead to less wasteful government spending, partly by encouraging yard-stick competition on the side of citizens and firms (e.g. Bordignon et al., 2004). Lastly, smaller jurisdictions are often argued to increase the quality of local democracy by promoting knowledge of local political issues and increasing local political efficacy and empowerment, leading to greater political participation. It is this last point which is the topic of our analysis. Political efficacy, which will be the main subject of analysis in the present paper, obviously is important in its own right if participatory democracy is valuable in itself (Almond & Verba, 1989: 136-139). Furthermore, however, the argument that decentralization improves service delivery is also critically dependent on a well-functioning local democracy, for two reasons. First, political participation has representational consequences: preferencematching, or allocative efficiency, will improve only if citizens make their preferences known through the local political process. Second, participatory democracy is a

3

Should have large lit fn here: Ellickson, Alesina and co-authors, Gilbert and Picard (1996), Besley and

Coate, Lockwood.

2

prerequisite for increasing electoral control and accountability, the two factors argued to be necessary for decentralization to improve productive efficiency. The purpose of this paper is to evaluate empirically the claim that jurisdiction size matters in a causal sense for the quality of local democracy. As noted by Campbell (1969), administrative reforms often provide researchers in the social sciences with opportunities for estimating causal effects by generating natural experiments that exogenously determines, or at least significantly influences, the treatment assignment in a non-experimental setting (Lassen, 2005). We exploit a recent large-scale administrative reform of the Danish municipal structure, the so-called Structural Reform (SR). Due to this reform, an individual residing in a below-average size municipality suddenly experienced a dramatic increase in the size of her jurisdiction. We investigate the consequences of this exogenous shock to jurisdiction size for political efficacy. Basically, if people in consolidated municipalities change their perception of their own political efficacy, we will say that there is a causal effect of population size. Our empirical analysis is based on two individual level surveys. The first was carried out in 2003, well before the reform, the second in December 2007 and January 2008, almost a year after the actual implementation of the newly formed municipalities, and three years after the first municipal election. The reform process was initiated by the central government, and had the explicit aim of generating larger municipalities to reap economies of scale. The reform was, as we shall argue in detail below, largely exogenous from the viewpoint of individual municipalities and definitely so for individual citizens. It is the fact that the reform was imposed on municipalities by the central government combined with the fact not all municipalities were affected by the reform in the same way that makes it an

3

attractive setting for studying the causal effect of jurisdiction size on the quality of local democracy. The 32 municipalities which were not amalgamated enables us to keep constant national level trends in local political interest, which could well be affected due to both the public debate on the large-scale reform and the concurrent revision of the system of intergovernmental grants. Furthermore, the reform did not affect the number or the placement of polling places, which allows us to isolate population size effects on voter turnout from spatial effects (see Brady and McNulty, 2004; Brady and Hui, 2006). We estimate the causal effect of jurisdiction size on local political efficacy using survey data on local political efficacy collected before and after the reform in both unaffected (control) and affected (treatment) municipalities. We find, using various differences-in-differences and matching estimators on both repeated cross-sectional data and retrospective evaluations, that citizens in amalgamated municipalities report lower political efficacy after the reform, and that the decline is larger for respondents coming from smaller pre-reform jurisdictions. The paper is structured as follows: The next section provides a theoretical framework for thinking about size, political participation, and selection. Section three describes the Danish structural reform, section four the data and section five lays out our research design and empirical methodology. Section six presents our results and section seven concludes with a discussion.

II. Jurisdiction size and local democracy The question of how jurisdiction size affects the democratic process is one of the oldest in political science. It can be traced almost 2,500 years back to Plato and Aristotle who preferred smaller entities, large enough to be self-sufficient, but small enough to ensure 4

that citizens can know one another’s characters (Dahl & Tufte, 1973: 4-5). Plato and Aristotle did not have modern democracies in their minds, but the idea that jurisdiction size affects the opportunities of citizens for participating in the democratic process is still relevant. Jurisdiction size may affect many aspects of the democratic process. Dahl & Tufte (1973: 13-16) mention eight of them: Jurisdiction size may affect the opportunity for citizens to participate, voluntary compliance and the prevalence of common norms and values, homogeneity, conceptions of the common good, loyalty, emotional ties to society, possibilities for accurate communication, and leader responsiveness to citizens. So not only is “What constitutes a good democratic process” a complex question. The potential effects of jurisdiction size are also multidimensional (Kjær & Mouritzen, 2003). Many empirical studies have investigated the effects of jurisdiction size and consolidation. Morland (1984) show that turnout is inversely related to jurisdiction size, as turnout is higher in national elections than in local elections and higher in smaller municipalities. Also other aspects of participation than turnout are affected by jurisdiction size (Verba & Nie, 1972). For instance, Oliver (2000) finds that voters in smaller jurisdictions are more likely to contact officials and to attend community or organizational meetings. High turnout and participation generally is an indication that citizens find it worthwhile to engage in politics, and is related to the perception of citizens of their own competence to participate in politics. This perception, political efficacy, is correlated to the tendency to participate in politics (Stenner-Day & Fischle, 1992). The concept of political efficacy was introduced by Campbell, Gurin & Miller (1954). They define it as the “feeling that individual political action does have, or

5

can have, an impact upon the political process, i.e., that it is worthwhile to perform one's civic duties.” The sense of high political efficacy is associated with more exposure to political communication and with higher satisfaction with participation in the political process citizens (Almond & Verba, 1989: 198, 193). Since these classic studies, the concept of political efficacy has been refined. Niemi, Craig & Mattei (1991: 1407) distinguish between two different dimensions of the concept. Internal efficacy refers to “beliefs about one’s own competence to understand, and to participate effectively in, politics” and external efficacy to “beliefs about the responsiveness of governmental authorities and institutions to citizen demands”. These dimensions have been detected in many studies of political efficacy in the US (for an overview see Niemi, Craig & Mattei, 1991: 1408), but also in comparative studies (Hayes & Bean, 1993). Empirical studies of the causes of individual citizens’ sense of internal political efficacy show that several factors, such as gender, age, education and income, are important (Almond & Verba, 1989; Hayes & Bean, 1993: 269-270). Jurisdiction is also found to affect efficacy, but there is some variation in the empirical results. The sense of internal political efficacy is stronger towards local government than towards the national government (Almond & Verba, 1989: 141; Dahl & Tufte, 1973: 57). As Dahl & Tufte’s eight aspects (those mentioned above) suggest, there are good reasons to expect similar differences between local units of varying size. Dahl (1970: 158) even argues that “For most citizens, participation in governing megalopolis must necessarily be largely symbolic, as in the realm of a prince. To be sure, the prince may be benign. (…) But so long as he wishes to act like a prince, only a few people can participate much in his decisions”. Almond & Verba (1963: 234) conclude that political efficacy

6

does not depend on city size. Finifter and Abrahamson (1975) argue that this conclusion is misleading. Since education levels tend to be higher in larger cities, and since education is positively related to the sense of political efficacy, the true relationship between city size and political efficacy may be suppressed. When education is controlled for, it is revealed that city size is in fact negatively related to feelings of internal political efficacy (Finifter, 1970: 403-404; Finifter & Abramson, 1975: 194). To sum up, many cross sectional studies investigate the democratic effects of jurisdictions size, and there is some, but not unequivocal, evidence that larger size is associated with various democratic problems. These studies are, however, plagued with several thorny methodological problems. First, differences in observable population characteristics can affect estimates (see Finifter and Abramson, 1975) and result in problems of common support rarely addressed in such studies. The population of small and large jurisdictions can be very different. Education is one important example (this was, according to Finifter & Abramson (1975), the root of the problem in Almond & Verba’s analyses of the relationship between city size and political efficacy). But also income levels and job types are different in small and large jurisdictions. It is hard to know whether differences in correlations are caused by jurisdiction size or simply by differences in the populations. This problem can, however, partly be solved by statistical control for the confounding variables. Second, important covariates at the jurisdiction level can be difficult to distinguish from jurisdiction size. Third, and most importantly, cross sectional studies cannot address problems of selection that arise if people with unobservable preferences over local

7

democracy self-select into jurisdictions of different size, partly with an eye on exactly this. The fact that people may sort themselves into different jurisdiction is precisely the Tiebout idea (Ellickson, 1971; Ross and Yinger, 1999). While there is in general little empirical support to the hypothesis that people sort based on income (e.g. Rhode and Strumpf, 2003), solid evidence for sorting based on preferences for local public goods such as school quality and environmental quality is beginning to emerge (Banzhaf and Walsh, 2008). Such sorting is also found in the Danish case analyzed below (Kristensen, 2002). This constitutes a problem of endogeneity. When such sorting effects exist, differences in political efficacy among jurisdictions of varying sizes may as much be the result of a reverse causal effect. This problem is hard to handle statistically. One solution is to use an instrumental variable approach, but perfect instruments are typically hard to find, and we have not been able to find any studies that use this approach to solve the problem. Another solution could by to study jurisdiction size and the quality of local democracy over time. This is, unfortunately, often infeasible, as important jurisdictional boundaries change only rarely. When change is observed in connection to large-scale reform, two additional complications arise: First, in large-scale reforms several things often change at the same time. For example, in reforms aimed at increasing citizen empowerment, decentralization (a decrease in the size of jurisdictions) can be accompanied by increasing transparency, changes in local political institutions and increases in funds from other levels of government, making it difficult to isolate the effects of changing jurisdiction size. Second, and related to the first, such reforms are often encompassing, leaving no individuals or jurisdictions unaffected.

8

In addition, when jurisdictional change happens, or is refrained from, it often does so endogenously. In many settings, jurisdictional change, whether at the city level (see Lassen, 2005), county level (see Alesina, Baqir and Hoxby, 2004) or country level (see Bolton and Roland, 1997, and Alesina and Spolaore, 1997) is often decided by voting. Alesina et al. (2004) finds that jurisdictions in the U.S. forego consolidation, by voting against it, if it means increasing population heterogeneity. Such majority consent is necessary for consolidation in many countries and settings, and introduces the issue of selection bias into estimates of the effect of jurisdiction size on local democracy indicators. Therefore, in order to identify a causal effect from the size of the populace to political participation it is necessary either to allocate people randomly across jurisdictions of different population size, which is something we do not do in this paper, or to identify a natural experiment which provides an exogenous source of variation in jurisdiction population size, which is what we do. In the next section we introduce the amalgamations of Danish municipalities, and argue that this provides an excellent opportunity for testing the effect of jurisdiction size on political efficacy. Before turning to the methodological considerations, we will introduce the Danish municipalities and give a short review of the existing Danish studies of the effects of jurisdiction size. The local level is a vital part of the Danish public sector. Danish local governments are responsible for about 40 per cent of Danish public expenditure, and relatively autonomous. They are free to set the rate of the local income tax and have wide discretion in most policy areas. Most notably, they control many important welfare services such as the school system, care for children and care for the elderly. The municipalities are governed by a city council elected by referendum for four year terms.

9

The mayor is elected indirectly by the city council. Apart from presiding over the council and being the formal head of the municipality, the mayor is the only full-time politician, head of the local administration, and chairman of the economic committee which assumes a central role in the political process (see Berg & Kjær, 2005). The local party systems are typically dominated by local branches of the national parties. Just as in Almond & Verba’s results, the sense of political efficacy in Denmark tends to be higher towards the local level than towards the national level and the European Union, although the differences between the local and the national level are limited (Andersen, 2000: 128-130). In a thorough cross-sectional study of the effect of jurisdiction size on internal political efficacy in Danish municipalities, Lolle (2003) concludes that jurisdiction size does not have a strong effect on political efficacy. The association between characteristics of citizens and their sense of political efficacy is much stronger. Especially education is strongly correlated with internal political efficacy (Andersen, 2000: 141-142; Andersen, 2003: 116-118), but Lolle (2003: 168) finds that the effect of education does not seem to depend on jurisdiction size. Another study of Danish municipalities showed a slight tendency for education to matter more in larger jurisdictions (Serritzlew 2004: 8). The effect of education on efficacy is, however, small at the local government level, and higher at the level of national government and of the European Union (Andersen, 2000: 144).

III. The Danish Structural Reform The Structural Reform (SR) in January 2007 dramatically consolidated Danish local government. The SR has three parts. First, the number of municipalities was reduced

10

from 2714 to 98. The average jurisdiction size increased from 20,100 to 55,600 citizens. Second, 14 counties were merged into five. Third, new tasks were to be undertaken by the municipalities. Due to SR, many individuals experienced a sudden increase in jurisdiction size. We make use of this exogenously induced change to estimate the causal effect of jurisdiction size on political participation. The SR was planned and implemented by the central government. After a short debate predominantly in newspapers on the possibility of reform, the government in 2002 appointed the Commission on Administrative Structure with a chairman appointed by the Minister for the Interior, three independent expert members, four representatives from municipalities and counties, and four representatives from Danish ministries. No publicly elected officials participated in this work. The four volume report of the committee was published in January 2004. It recommended a large scale structural reform. Based on this, the minority government introduced a bill, which immediately got a majority when supporting Danish People’s Party declared that they had one question for the negotiations with the government: Where shall we sign? After less than half a year, in June 2004, it was clear that Danish local government should experience dramatic reform. According to the agreement between the government and Danish People’s Party, the municipalities should be merged into larger units with at least 30,000 citizens. All municipalities with less than 30,000 citizens (230 of the 270 municipalities, see Table 1) were asked to arrange for mergers before January 1st 2005. Only neighbouring municipalities would be allowed to merge. Eleven of the new 98 municipalities had 4

When the debate over the reform began in 2002, five smaller municipalities on the island of Bornholm

voluntarily merge into one island-wide municipality, reducing the total number of municipalities from 275 to 271.

11

already before this agreement announced how they wished to merge, and most of the remaining municipalities were very close to agreeing on with whom to merge (BlomHansen et al., 2006: 17-8). Some municipalities, especially islands, would be allowed to continue with less than 30,000 citizens. Such municipalities would have to cooperate with larger municipalities in providing some services to their citizens.

< Table 1 here >

Of the 270 original municipalities, 237 were merged into 65 new ones, and the remaining 33 municipalities continued without any changes. The first election involving the post-reform municipalities as political entities took place in November 2005, while the new structure came into effect from January 1st 2007. From this date the municipalities were made responsible for additional tasks (most importantly certain social services for handicapped, prevention, rehabilitation, and special education) amounting to almost DKR 25 bill. The new responsibilities are, however, minor compared to the total municipal expenditure of (in 2007) DKR 333 bill. The main problem of studying effects of jurisdiction size by comparing existing entities with varying population size is sorting effects, where individuals’ choice of municipality depends on something related to the dependent variable of interest. By studying on a natural experiment with exogenously induced change in jurisdiction size, this problem is minimized. There are, however, still three sources of endogeneity. First, individuals can adapt to the new municipal structure by moving. Individuals with strong preferences for political participation at the local level may choose to move to one of the small municipalities. Since we have conducted the survey

12

within a year after the reform, this problem is quite small. We compensate for it by controlling for whether an individual has moved since the reform. Second, although individuals were not allowed to choose if and how their municipality should be merged, the municipalities did have substantial influence on this. Municipalities of citizens with a strong preference for local democracy may have aimed for a merger with a small municipality. We regard this a minor problem as the options of the municipalities were constrained by the size requirements and the condition that merging municipalities should be neighbours. Third, referenda were held in 63 of the original 270 municipalities (Jørgensen, 2006: 168). Most of them concerned adjustments of boundaries, but some involved the choice between a merger with one or another municipality. In our sample of 32 municipalities, referenda took place in six cases. They are marked with asterisks in Table A in the appendix. In these cases citizens did have a direct influence on what would happen to their municipality. Individuals did, of course, only have a minuscule chance of affecting the result.

IV. Data We use two individual level surveys, carried out by phone by a professional polling firm, on voter participation in local democracy. One was carried out before the SR, in 2001, the other was carried out in late 2007 and early 2008, a year after the SR was administratively implemented. A core set of questions were asked in both surveys, making it possible to compare pre- and post-SR political behaviour. The 2001-survey was a sample of 4,700 residents in Denmark aged 18 or more. The overall response rate was 59 percent. Respondents were randomly selected in 60 of the then 275 Danish municipalities. The 60 municipalities were randomly selected 13

from six strata based on size, with ten municipalities from each (see Houlberg & Strømbæk Pedersen, 2003). The 2007-survey was a sample of 2,816 residents in Denmark. Of these, 709 either was younger than 24 years old, had non-functional phone numbers, or could not be reached in at least six attempts. Of 2107 contacted respondents, 59 percent completed the survey. Since the aim is to compare the two surveys, only residents aged 24 (who were 18 when the first survey was carried out) or more were included. Younger respondents were excluded in order rule out any cohort effects. Only municipalities selected in the 2001-survey were included in the 2007-survey. The 2007- survey was stratified by pre-SR municipalities, making it possible to compare municipal level responses in pre-SR municipalities within the same area, now part of a new, larger postreform municipality. At the same time, we also surveyed municipalities unaffected by the reform. The 60 municipalities of the 2001-survey were divided into five strata. Table 2 shows the strata and the 32 selected municipalities.

< Table 2 here >

Eight were randomly selected from the first stratum, and six from each of the remaining four strata. The first stratum consists of municipalities which were not merged during the reform. The second stratum consists of merged municipalities with jurisdiction size below medium and a relative increase in size above medium. The second stratum consists of municipalities with below medium size and above medium growth, the third of municipalities with below medium size and below medium growth,

14

the fourth of municipalities with above medium size and above medium growth, and finally the fifth of municipalities with above medium size and growth. The selected municipalities are listed in Table A in the appendix.

< Table 3 here >

We measure internal political efficacy by the five standard questions described in Niemi, Craig & Mattei (1991). These questions are used in the U.S. National Election Study. The questions has been translated into Danish and adapted to local politics. The (English translation) of the questions appear in Table 3. We use the exact same formulation in the 2007-survey as was used in the 2001-survey (see Lolle, 2003). Table 3 also shows descriptive statistics and definitions of the internment variables used in the analyses below.

V. Research design and empirical specification Our basic research design is to estimate the causal effect of jurisdiction size on internal political efficacy using the differences-in-differences (DiD) method on repeated cross sectional data. DiD is well known in the program evaluation literature (see, e.g., Imbens and Wooldridge (2008) for a recent treatment and Finkelstein (2002) for a straightforward application), and describes the situation in which some units or individuals receiving treatment are compared to their pre-treatment levels, while the

15

same is done for a control group. This yields the first level of differences.5 Subsequently, these two differences are subtracted from each other, which is the second level of differences, leaving an estimate of the causal effect of treatment under certain assumptions.

V.1. THE DIFFERENCES-IN-DIFFERENCES ESTIMATOR Let the internal efficacy of an individual i be Yi (1) in the treated case, that of municipal amalgamation, and Yi ( 0 ) be the outcome for the same person in case of no treatment. Our object of interest is the difference Δ i = Yi (1) − Yi ( 0 ) which is the effect of municipal consolidation on the internal political efficacy experienced by individual i. In practice, however, an individual i cannot be treated and non-treated at the same time, which means that either Yi (1) or Yi ( 0 ) will be missing. The difference Δ i results from the pair of potential outcomes, a term coined by Rubin (1974), which is to be distinguished from the realized outcome, Yi . Since a counterfactual for the realized outcome is not available at the individual level, the statistical approach is to estimate the missing variable from appropriate group means. We study a repeated cross section model of individuals surveyed in 2001, well before the reform took place, and individuals surveyed in December 2007, two years after the first election of the newly formed municipalities and almost one year after the actual implementation. This model has the form

Yi = α + γ M M i + γ PTi + τ DID M iTi + β ' X i + ui ∀i ∈ I ,

5

In the early evaluation literature, this is also known as “interrupted time series with a nonequivalent no-

treatment control group time series”-design. See Cook and Campbell (1979, p. 214).

16

where I is the set of respondents. M i ∈ {0,1} is an indicator for reform (as opposed to continuing) municipalities. Below, we also consider a treatment indicator of the form M i ∈ {0, SS , SL, LS , LL} where we distinguish treated municipalities both by their pre-

reform size and their post-reform relative size, to deal with unobserved heterogeneity; we return to this below. Ti ∈ {0,1} is a time period indicator equal to 0 before the reform and 1 after the reform. The interaction(s) M iTi take(s) on the value 1 for reform municipalities after implementation, and zero otherwise. Finally, ui represents unobservable characteristics, and is assumed to be independent of treatment status conditional on observables and across periods; we return critically to this assumption below. It can be useful to consider the conditional means for the four different groups: reform and non-reform municipalities, before and after reform. They are as follows: (A) Y00 = E (Y | M = 0, T = 0, X ) = α + β ' X (B) Y01 = E (Y | M = 0, T = 1, X ) = α + β ' X + γ P

(1)

(C) Y10 = E (Y | M = 1, T = 0, X ) = α + β ' X + γ M (D) Y11 = E (Y | M = 1, T = 1, X ) = α + β ' X + γ P + γ M + τ DID

The Before-After estimator for the Treated, BAT, used when no control group is available, is given by (1.D) – (1.C): Y11 − Y10 = γ P + τ DID

(2)

The BAT-estimator includes both the effect of treatment and the common trend or aggregate change. In our case, γ P includes the effects on internal political efficacy from reallocating responsibilities across levels of government, which are the same in consolidated and continuing municipalities alike. To estimate consistently the effect of

17

consolidation, we need to subtract this common trend. This is done by subtracting the Before-After estimator of the Control group, BAC:

Y01 − Y00 = γ P .

(3)

The BAC estimates the effect of reallocation of government tasks as well as any aggregate time trends in internal political efficacy with respect to local government; the latter may have increase or decreased in reaction to the substantial public debate on municipal politics and the Structural Reform. Similarly, increased fiscal pressure from central authorities on municipalities will also be captured by this trend, as long as it is similar across all municipalities. Based on our set-up in (1), we get the difference-in-difference estimator as

τ DID = (Y11 − Y10 ) − (Y01 − Y00 )

(4)

= (Y11 − Y01 ) − (Y10 − Y00 )

where the first row is the BAT minus the BAC estimator, while the second row shows that we can also interpret the DiD-estimator as the difference between the two crosssectional estimators. The assumption that the trend in (2) is common for treatment and control groups alike is a key identifying assumption in the DiD-approach (see Blundell and Macurdy, 1999), and we see no reason why it should not hold in our application; as noted above, the reallocation of government tasks across vertical levels of government was the same for all municipalities. However, if the impact of the reform is heterogeneous with respect to observable characteristics, and the distributions such observable characteristics differ across control and treatment groups, additional assumptions are needed to make sure a comparison is based on suitably similar groups; we return to this below, where we consider matching. The second key identifying assumption, denoted 18

no composition bias, is that the populations considered are the same across time. At the population level, this means that no individuals change from treatment to control group and vice versa; this is sometimes problematic in studies of individual responses to changes in taxation or job training programs, but in our case this would be relevant only if people moved with the explicit aim of living in a treatment or control municipality. This seems highly unlikely. In addition, in our empirical analysis we exclude people in the post-reform sample that recently moved from another municipality. The assumption is more problematic when we consider the actual samples of individuals; we return to this below. When estimating individual level responses to aggregate level variables, here treatment status and other municipal-level variables, it is important to account for the possible covariance in errors. We do this by allowing for clustering at the (new) municipal level; this specification allows for even higher correlation in sub-clusters such as old municipal levels. Our number of sampled clusters is not sufficiently large for us to be able to refer to asymptotic results. Therefore, in the table below we report bootstrapped standard errors, resampled on clusters, and report bias-corrected confidence intervals (see Donald and Lang, 2007, and Cameron, Gelbach and Miller, 2008).6

V.2. NON-RANDOM ASSIGNMENT: DIFFERENTIAL SURVEYS AND DIFFERENCES IN OBSERVABLES ACROSS TREATMENT GROUPS

6

This is essentially the same as using multi-level modeling. However, here we refrain from parametric

assumptions, both for the OLS which is a semi-parametric estimator, and in matching, which is nonparametric.

19

In existing DiD-studies on repeated cross sectional data, the analysis is typically carried out as different waves of a particular representative survey, such as the Canadian GSS in Finkelstein (2002). In our case, we compare internal political efficacy measures across surveys carried out in the same municipalities, using exact similar worded questions, but with different sampling strategies. The different sampling probabilities raise the problem that the surveys are not stratified in the same way, and thus it is possible that controlling for confounding variables does not solve the problem of composition bias. As noted above, an assumption of no composition bias is necessary for identification of a DiD-model (Blundell and Macurdy, p. 1612). To solve this problem, we want to match respondents across surveys on a number of individual level variables and their interactions. At the same time, as noted above, treatment and control municipalities may differ with respect to observable characteristics, which can lead to a violation of the common trend assumption if effects of the reform are heterogeneous across observables; as noted above, for example, an explicit aim of the SR was to create municipalities with a minimum of 30,000 inhabitants, which in itself affects the probability of treatment.7 To solve this issue, we want to match survey respondents across treatment status. In sum, therefore, there are two assignments that may be nonrandom: assignment to treatment or control group and assignment to pre-reform or postreform survey. To address this, we employ a differences-in-differences matching

7

This is an example of the criterion, stressed by Rubin (2008, p. 820), of including explicitly the reasons

for treatment assignment at the level of the decision-maker.

20

(DiDM) estimator, which was developed in the program evaluation literature by Heckman et al. (1997, 1998). 8 We proceed by propensity score matching. 9 While intuitively desirable, exact matching on a vector of characteristics is often infeasible due to the large number of possible combinations. As shown by Rosenbaum and Rubin (1983), however, if potential outcomes are independent of treatment conditional on X, which is a precondition for estimating a causal effect from the data anyway, then this is also the case conditional on the propensity score, defined as the probability of receiving treatment conditional on X. As we have potentially two non-random assignments, we follow Blundell et al. (2004) in defining two propensity scores. Conditional on X, PM = Pr ( M = 1| X ) is the probability of being observed in a treated municipality and PT = Pr (T = 1| X ) is the probability of being surveyed in the post-reform sample. Based on the propensity scores, we can now write the identifying assumption as E (Yit0 | PM , PT , M = 1, T = 1) − E (Yit0 | PM , PT , M = 1, T = 0 ) = E (Yit0 | PM , PT , M = 0, T = 1) − E (Yit0 | PM , PT , M = 0, T = 0 ) . This allows the time effects to differ by X and implies that the distribution of observed characteristics is the same across all four samples. In practice, we estimate the two propensity scores by running a probit (or logit) regression of assignment status (treatment and time, respectively) on X. The resulting

8

Imbens and Wooldridge (2008) provide a recent overview of development on program evaluation,

covering both matching and selection on unobservables, including the DiD-estimator. 9

Imai (2005) provides an excellent introduction to propensity score matching in a political science

setting. Barabas (2004) and Lassen (2005) use it to evaluate effects of deliberation based on observational data (check this) and as a robustness check in a natural experiment, respectively.

21

predicted values are the estimated propensity scores. Figure 1 shows the distributions of propensity scores across cells. Our point of departure is the group of individuals in the treated group, pre-treatment (M = 1, T = 0). To match these individuals with respondents in each of the three control groups, we perform a variant of nearest neighbor caliper matching with replacement by pairing each treated respondent with a respondent in each of three control groups. The respondent in each control group is chosen to minimize the Euclidean distance from the treated individual’s two estimated propensity scores, given restrictions on the maximum allowable distance (the caliper) both for the two-dimensional distance measure and for the two one-dimensional propensity scores. Based on these matched samples, we carry out both simple differences-in-differences analysis and nonparametric matching to form estimates of the ATT under matched DiD under the assumption of separable additivity of time and group effects. Since matching is carried out with replacement, we correct the sample sizes to reflect the weights implied by this procedure, but show results for both cases. Throughout, we, as is standard in the literature, report bootstrapped standard errors for the estimated ATTs based on the DiDM procedure. A recent study shows that this may in fact result in biased estimates for the standard error, but the issue is unsolved (Abadie and Imbens, 2007). In performing the bootstrap, we resample at the cluster level of new municipalities (see above).

V.3. UNOBSERVED HETEROGENEITY

While matching improves upon or sometimes even eliminates problems arising from non-random assignment on observable variables, it does not address the potential confounding of community size with unobserved characteristics of individuals with

22

respect to attitudes to local democracy. This can happen if people self select into municipalities of different size, possibly partly as a function of interest in or preference for local democracy. If unobserved heterogeneity is present and correlated with treatment status, the DiD-estimator is miss specified, since the expected error term will be present in the conditional means in (1) and, thus, in the resulting estimators.10 To deal with these issues, we consider exact matching based on the categorical treatment variables (SS/SL/LS/LL) as well as on pre-reform municipalities. This means that matching on the Euclidean distance is carried out under the restriction that the matched observation in the pre-reform treatment group must be found in the same category of municipalities or, for the alternative case considered, in the same pre-reform municipality.11 As emphasized by Heckman et al. (1997) and Smith and Todd (2005) in their work on causal effects of job training programs, and confirmed more generally by Cook, Shadish and Wong (2008), DiDM functions best when matching individuals who reside within a given geographical unit and, if possible, are administered identical surveys. This is exactly what we do here.

V.4. RETROSPECTIVE EVALUATIONS

To supplement the evidence from the repeated cross sections, we also asked survey respondents in the 2007 survey to assess, in retrospect, the consequences of the

10

Morton and Williams (2008, chapter 10) provide an introduction to the structural approach to causality

and its focus on unobserved heterogeneity in a political science context. As noted above, the DiDestimator is essentially structural, as noted by Blundell and Macurdy, even if most applications of it in a natural experiment context present it as a non-structural approach. 11

An alternative would be to treat each of the municipal treatment categories, or even the municipalities

themselves as separate treatments in the DiDM-analysis, but sample size makes this infeasible.

23

Structural Reform for measures of local political efficacy. While there are well-known issues with retrospective questions of this kind,12 they allow us to get at the missing panel data in a different way: by asking one person to evaluate the differences before and after the reform, we essentially control for individual level factors and get a direct estimate of the first level differences. We can now compare these differences across treatment and control municipalities as well as within the group of treatment municipalities. We analyze these data using both a regression-based cross-sectional framework and matching across treatment and control groups, for both the full sample and the matched sample resulting from the DiDM-analysis. Throughout we take into account the clustered (or multi-level) nature of the data by allowing for within-municipality dependence in estimated standard errors.

VI. Results VI.1. REGRESSION-BASED DID-ESTIMATES

Table 4 reports results from the standard, regression-based difference-in-difference analysis, for three different specifications: analysis of binary, categorical and continuous treatments on the full samples.

< Table 4 here >

12

These questions were purposely asked after the standard efficacy level questions, so as not to prime

respondents.

24

The results for the binary treatment, where citizens in all reform municipalities are considered identically treated, suggest strong effects of the reform. First, the estimate on the post-reform variable suggests a general significant and negative effect of the entire reform package on all municipalities (but we return to this below). Second, there seems to be no significant differences across treatment and control municipalities before the reform. Turning to the ATT, the effect of the reform on citizen political efficacy in treated municipalities, we estimate the difference-in-difference effect to be -.77 for the full sample, bringing the total effect of reform for the treated municipalities to approximately -1.3. The DiD-estimate is significant at the 5%-level.13 Column 2 presents results from the categorical treatment. In this case, we distinguish municipalities by the pre-reform size and their post-reform relative size, creating the four categories SS, SL, LS and LL presented in table 2 above. This way of looking at the data shows quite different results for the four treatment categories: Respondents living in pre-reform small municipalities that post-reform are a relatively small share of total new municipality size experience the largest decline in internal political efficacy, a decline which is significant at the 1 % level, while respondents in the three other categories experience almost equivalent declines in internal political efficacy. Finally, column 3 reports results from the continuous treatment, where we consider the effect of the change in population size experienced by respondents. Population size difference corresponds in a monotonic way to the categorical treatment variables used in column 2 and, indeed, we find results similar to those reported for the 13

As noted earlier, all standard errors are corrected for clustering at the new municipal level. Since the

number of clusters is not large, reported standard errors are the results of a bootstrap procedure with cluster-level resampling and 200 replications.

25

categorical case: The loss of internal political efficacy experienced by respondents is increasing in the population increase following municipal amalgamation.

VI.2. RESULTS FROM DIFFERENCES-IN-DIFFERENCES WITH MATCHING (DiDM)

Table 5 shows results from the matching analysis. The top panel reports results from the specification where treated individuals post-reform were matched to individuals in similar municipal categories, pre-reform. The left hand part of the top panel shows results from the matched, unweighted sample. The binary ATT is -.967 which is significant at the 5 percent level and a 25 percent increase relative to the estimate from the full-sample regression based result reported in table 4. As above, the categorical treatment regime suggests large difference across categories: The largest, and most precisely estimated, effect is for individuals coming from small pre-reform municipalities that constitute a small share of the post-reform municipality. The estimated ATT for these individuals is -1.9. The pattern generally is the same as in the regression-based analysis, with small and insignificant effects for citizens living in municipalities where there was little change ex post, as their pre-reform municipality remained the larger one.

< Table 5 here >

The right hand part of the top panel shows results when control unit observations are weighted by the number of times they are used for matching, since this took place with replacement. The estimated effects are larger than in the unweighted

26

case: the estimated ATT for the binary case increases by more than 25 percent. This suggests that matches in the control units that are used several times (ranging from two to, in one case, thirty replacements) report comparatively higher efficacy than the average for such units. In the lower part of the table, we show results when the matching of treated individuals across time was done under the additional restriction that a match should be found in the same pre-reform municipality, rather than municipal category. This means that a survey respondent in the post-reform treatment group living in a particular area of a larger amalgamated municipality should be matched to a respondent living in the same area, which pre-reform was a municipality of its own. This is a restrictive condition and, indeed, a binding restriction: the number of observations drop from 1,417 to 1,226, a decrease of 13 percent, as it is now impossible to find suitable matches for some post-reform treated respondents. For the unweighted case, the left hand side of the lower panel of table 5, the difference is not large; the main effect is that we observe a significant, but quite small, decrease in political efficacy reported by respondents in type LL-municipalities. When we turn to the weighted sample, however, the estimated effect is considerably larger than in the similar panel above. For the categorical case, we estimate an ATT in excess of 1 for all categories, strongly significant for 3 out of 4. Again, this results from the fact that some control observations are now sampled even more than in the weighted case above. The conclusion is clear: taking into account non-random assignment across samples, in both the treatment and the time dimension, increases the estimated effects of municipal amalgamation considerably. The estimated ATT from the weighted sample with exact matching on pre-reform municipalities (equal to 1.608) is, almost

27

exactly, double that estimated by the simple regression-based approach from table 4 (equal to .772). Furthermore, in contrast to the standard regression-based results, the DiDM-analysis suggests no overall change in political efficacy as a consequence of the reform. Thus, for all results reported in table 5 the coefficients on the post-reform period (T = 1), as is the case for the coefficients on the treatment group (M = 1), are always insignificant (not shown). This suggests that composition effects, owing to differences in survey respondent profiles, may be responsible for the apparent decline in countrywide internal political efficacy at the local level observed above, and this confirms the importance of addressing non-random selection both across treatment status and across surveys. Finally, in results not reported in table 5, we find that estimating the direct effect of the continuous treatment, the actual change in population size, on the matched sample yields estimates that are approximately 50 percent larger than the regressionbased estimates presented in table 4. The point estimate on the continuous treatment effect is -.232 (p-value = .002) when population differences are measured in 10,000s.

VI. 3. RETROSPECTIVE EVALUATIONS

Table 6 shows the results from the retrospective questions. Each cell in the table shows the result of an estimation process, either in the form of linear regression or, for the bottom row, propensity score kernel matching. All estimations, including estimation of the propensity score, include a full set of controls as above.

< Table 6 here >

28

For all estimation methods and samples we consistently find a statistically significant decline in perceived political efficacy for individuals in reform municipalities. The results on the binary treatment indicator, reported in the top row, suggest reasonably similar effects across samples, equivalent to one half standard deviation. Identical results are obtained using kernel matching, reported in the bottom row. The effect is not homogenous across treatment categories, however: The impact of reform is larger for individuals residing in small pre-reform municipalities and generally small and insignificant for inhabitants of the largest reform jurisdictions, when these jurisdictions constitute a large part of the newly formed municipality. Throughout, F-tests reject the null hypothesis of equal parameter estimates for the four categories. These results are confirmed by the continuous treatment variables: The effect of amalgamation is larger, the larger is the change in population size resulting from the reform; thus, individuals experiencing larger (both absolute and relative) changes in the size of their jurisdiction reported greater losses of perceived political efficacy.

29

VII Conclusion

The structure of government matters for economic as well as political outcomes. The optimal size of a polity is not easy to identify, as it involves trade offs between factors such as economies of scale and allocative efficiency, and democratic participation. Smaller jurisdictions may increase the quality of local democracy by promoting knowledge of local political issues and increasing local political efficacy and empowerment, leading to greater political participation. In this paper we focus on the effects of jurisdiction size on one important aspect of political participation and democracy: Internal political efficacy. This relationship has been investigated in cross sectional studies, but the question is still unresolved. Almond & Verba (1963: 234), for example, conclude that political efficacy does not depend on city size, but Finifter and Abrahamson (1975) argue that this is misleading. Since education levels tend to be higher in larger cities, and since education is positively related to the sense of political efficacy, the true relationship between city size and political efficacy may be suppressed, and they find that when this is taken into account, jurisdiction size appear to be negatively related to feelings of internal political efficacy. The cross sectional approach is generally faced with three difficult methodological problems. Differences in observable population characteristics can affect estimates, important covariates at the jurisdiction level can be difficult to distinguish from jurisdiction size, and cross sectional studies cannot address problems of selection that arise if people with unobservable preferences over local democracy self-select into jurisdictions of different size. 30

We address these problems by identifying a natural experiment which provides an exogenous source of variation in jurisdiction population size. Due to the Danish structural reform of local government, individuals residing in a below-average size municipality suddenly experienced a dramatic increase in the size of her jurisdiction. We investigate the consequences of this exogenous shock to jurisdiction size for a widely used measure of political participation and engagement, the individual internal political efficacy. Standard regression-based difference-in-difference analysis shows considerable and statistically significant detrimental effects on political efficacy for treated municipalities. The loss of internal political efficacy experienced by respondents is increasing in the population increase following municipal amalgamation. In matching analyses non-random assignment across samples in both the treatment and the time dimension is taken into account. In this specification we find even stronger effects of jurisdiction size. Furthermore, and in contrast to the regression-based results, the matching-analysis suggests no overall change in political efficacy as a consequence of the reform. Finally, results from the retrospective questions also consistently show that perceived political efficacy is lower when jurisdiction size increases. We conclude that there is strong evidence that jurisdiction size has a causal effect on internal political efficacy. Smaller polities do in fact offer a better environment for this important aspect of political participation and democracy. Finally, we note that in addition to the effects of jurisdictional change being different across polities, we observe substantial treatment heterogeneity across individuals. In ongoing work, we find that the effect of reform is large and significant for men, but we find no effect for women. In a similar way, there seems to be little

31

effect of the reform on the political efficacy for the young and the old. One potential reason is that women, as well as the young and the old, pay less attention to local politics and therefore are less likely to observe both positive and negative changes in the political setting. As a result, centralization may in fact reduce, rather than enhance, inequality in political participation.

32

References Alesina, Alberto and Enrico Spolaore. 1997. "On the Number and Size of Nations." Quarterly Journal of Economics 112. Alesina, Alberto and Enrico Spolaore. 2003. Book on nations. Alesina, Baqir and Hoxby. 2004. "Political Jurisdictions in Heterogeneous Communities." Journal of Political Economy 112:348-396. Almond, Gabriel A. & Sidney Verba. (1963). The Civic Culture. Princeton, N.J.: Princeton University Press Almond, Gabriel A. & Sidney Verba. (1989). The Civic Culture. Princeton, N.J.: Princeton University Press Andersen, Jørgen Goul (2000). ”Magt og afmagt: Nyt perspektiv på ’political efficacy’”, pp. 123-152 in Jørgen Goul Andersen, Lars Torpe & Johannes Andersen (eds.) Hvad folket magter. Demokrati, magt og afmagt, Købehavn: Jurist- og Økonomforbundets Forlag Andersen, Jørgen Goul (2003). Over-Danmark og under-Danmark? Ulighed, velfærdsstat og politisk medborgerskab. Århus: Aarhus Universitetsforlag. Banzhaf and Walsh. 2008. “Do People Vote with Their Feet? An Empirical Test of Tiebout’s Mechanism.” AER. Barabas, J. “How Deliberation Affects Policy Opinions.” American Political Science Review 98, 687-701. Barankay, I. and B. Lockwood. 2007. “Decentralization and the productive efficiency of government: Evidence from Swiss cantons.” JPubE. Benton, J.E. and D. Gamble. 1983. “City/County Consolidation and Economies of Scale: Evidence from a Time-Series Analysis in Jacksonville, Florida.” Social Science Quarterly 65, 190-198. Berg, Rikke & Ulrik Kjær. 2005. Den danske borgmester, Odense: Syddansk Universitetsforlag. Besley, T. and S. Coate. 2003. “Central versus local provision of public goods: a political economy analysis.” Journal of Public Economics 87, 2611–2637. Blundell, R. and T. Macurdy. 1999. “Labor Supply: A Review of Alternative Approaches.” Chapter 27 in Handbook of Labor Economics Volume 3A, 15591695. 33

Blundell, R., M. Costa Dias, C. Meghir and J. Van Reenen. 2004. “Evaluating the Employment Impact of a Mandatory Job Search Program.” JEEA 2, 569-606. Bordignon, M. F. Cerniglia and R. Revelli. 2004. “Yardstick competition in intergovernmental relationships: theory and empirical predictions.” Economics Letters 83, 325–333. Brady and Hui, Polmeth meetings, 2006. Brady, McNulty, 2004. Cameron, A.C., J. Gelbach and D. Miller. 2008. “Bootstrap-based improvements for inference with clustered errors.” Review of Economics and Statistics 90, 414427. Campbell, Angus, Gerald Gurin & Warren E. Miller (1954). The Voter Decides. Evanston: Row, Peterson. Campbell, D.T. 1969. “Reforms as Experiments.” American Psychologist. Cook, T. and D. Campbell. 1978. Quasi-Experimentation. Cook, T., W. Shadish and V. Wong. 2008. “Three Conditions under Which Experiments and Observational Studies Produce Comparable Causal Estimates: New Findings from Within-Study Comparisons.” JPAM 27, 724-750. Craig, Stephen C., Richard G. Niemi & Glenn E. Silver (1990), “Political Efficacy and Trust: A Report on the NES Pilot Study Items”, Political Behavior, Vol. 12, No. 3, pp. 289-314. Dahl, Robert A. (1970). After the Revolution?. New Haven: Yale University Press. Dahl, Robert A. (1989). Democracy and its Critics. New Haven: York University Press. Dahl, Robert A. and Edward R. Tufte (1973). Size and Democracy. Stanford: Stanford University Press. Dehejia, R. and S. Wahba. 2002. “Propensity Score Matching Methods for Nonexperimental Causal Studies.” Review of Economics and Statistics 84, 151161. DeHoog, R., D. Lowery and W. Lyons. 1990. “Citizen Satisfaction with Local Governance: A Test of Individual, Jurisdictional, and City-Specific Explanations.” Journal of Politics 52, 807-837. Donald, S. and K. Lang. 2007. “Inference with differences-indifferences and other panel data.” Review of Economics and Statistics 89, 221-233. 34

Ellickson, Bryan. 1971. “Jurisdictional Fragmentation and Residential Choice.” American Economic Review 61, 334-9. Ellingsen, T. 1998. “Externalities vs Internalities: A Model of Political Integration.” JPubE. Faguet, Jean-Paul. 2004. "Does Decentralization Increase Responsiveness to Local Needs? Evidence from Bolivia." Journal of Public Economics 88, 867-894. Finifter, Ada W. & Paul R. Abramson (1975), “City size and Feelings of Political Competence”, Public Opinion Quarterly, vol. 39, no. 2, pp. 189-198 Finifter, Ada W. (1970), “Dimensions of Political Alienation”, American Political Science Review, vol. 64, no. 2, pp. 389-410 Hayes, Bernadette C. & Clive S. Bean (1993), “Political efficacy - a comparative-study of the United States, West Germany, Great Britain and Australia”, European Journal of Political Research, vol. 23, pp. 261-280. Heckman, J., H. Ichimura, J. Smith, P. Todd. 1998. “Characterizing Selection Bias Using Experimental Data.” Econometrica 66, 1017-1098. Heckman, J., H. Ichimura, P. Todd. 1997. “Matching as an Econometric Evaluation Estimator: Evidence from Evaluating a Job Training Programme.” Review of Economic Studies 64, 605-654. Hindriks, J. and B. Lockwood. 2005. “Decentralization and electoral accountability: incentives, separation and voter welfare.” CEPR Discussion Paper 5125. Hooghe and Marks. 2003. “Unraveling the Central State, but How? Types of MultiLevel Governance.” APSR. Imai, Kosuke. (2005) “Do Get-Out-The-Vote Calls Reduce Turnout? The Importance of Statistical Methods for Field Experiments.” American Political Science Review 99, 283-300. Imbens, G. and J. Wooldridge. 2008. “Recent Developments in the Econometrics of Program Evaluation.” NBER Working Paper #14251, August. Kjær, Ulrik & Poul Erik Mouritzen (2003), “Er der sammenhæng mellem kommunestørrelse og lokalt demokrati?”, in Kjær & Mouritzen (eds.) Kommunestørrelse og lokalt demokrati, Odense: Syddansk Universitetsforlag

35

Kristensen, Thomas Balle (2002), ” Kommunestrukturens konsekvenser - politik, økonomi eller politisk økonomi?”, in Blom-Hansen, J., Bruun, F., Pallesen, T. (eds.) Kommunale patologier, Systime, Århus. Lassen, David Dreyer. 2005. “The Effect of Information on Voter Turnout: Evidence from a Natural Expriment.” AJPS. Lolle, Henrik (2003), “Kommunalpolitisk selvtillid”, pp. 162-176 in Ulrik Kjær & Poul Erik Mouritzen (eds.), Kommunestørrelse og lokalt demokrati, Odense: Syddansk Universitetsforlag. Lowery et al. APSR 1995. Lyons, W. and D. Lowery. 1989. “Governmental Fragmentation versus Consolidation: Five Public-Choice Myths about How to Create Informed, Involved, and Happy Citizens.” Public Administration Review 49, 533-543. Morlan, R. 1984. “Municipal vs. National Election Voter Turnout: Europe and the United States.” Political Science Quarterly 99, 457-470. Morton, R. and K. Williams. 2008. From Nature to the Lab: Experimental Political Science and the Study of Causality. Book manuscript, NYU, September 15. Niemi, Richard G., Stephen C. Craig & Franco Mattei (1991), “Measuring Internal Efficacy in the 1988 National Election Study”, American Political Science Review, vol. 85, no. 4, pp. 1407-1413. Oberholzer-Gee & Strumpf. JPE. Oliver, Democracy in Suburbia, Princeton UP. Oliver, J. Eric. 2000. "City Size and Civic Involvement in Metropolitan America." American Political Science Review 94:361-73. Rhode and Strumpf. 2003. “Assessing the Importance of Tiebout Sorting: Local Heterogeneity from 1850 to 1990.” AER. Rogers, B.D. and C.M. Lipsey. 1974. “Metropolitan Reform: Citizen Evaluations of Performances in Nashville-Davidson County, Tennessee.” Publius 4, 19-34. Rosenbaum and Rubin. 1983. Biometrica. Ross, Stephen and John Yinger. 1999. “Sorting and Voting: A Review of the Literature on Urban Public Finance.” Chapter 47 in Handbook of Regional and Urban Economics, Amsterdam: Elsevier.

36

Rubin, D. “For Objective Causal Inference, Design Trumps Analysis.” Annals of Applied Statistics 2, 808-840. Seabright, P. 1996. “Accountability and decentralization in government: an incomplete contracts model.” European Economic Review 40, 61–91. Serritzlew, Søren (2004), ”Det lokale demokrati i den nye kommunestruktur”, FokusNyt, no. 27, pp. 4-9. Smith, J. and P. Todd. 2005. “Does matching overcome LaLonde’s critique of nonexperimental estimators?” Journal of Econometrics 125, 305-353. Stenner-Day, Karen & Mark Fischle (1992), “The Effects of Political Participation on Political Efficacy: A Simultaneous Equations Model”, Australian Journal of Political Science, vol. 27, pp. 282-305. Tiebout, C. 1956. “A Pure Theory of Local Expenditures.” JPE. Treisman, D. 2007. The Architecture of Government: Rethinking Political Decentralization. Cambridge University Press. World Bank. 2000. “Entering the 21st Century” in World Development Report 2000, Washington D.C.

37

Table 1: Number of Danish municipalities

Number of citizens

Municipalities in 2006

Municipalities in 2007

Frequency

Percent

Frequency

Percent

0-5,000

14

5

3

3

5,001-10,000

114

42

1

1

10,001-20,000

77

29

3

3

20,001-30,000

25

9

18

18

30,001-50,000

23

9

39

40

50,001-100,000

13

5

28

29

More than 100,000

4

1

6

6

Total

270

100

98

100

Source: Statistics Denmark.

Table 2: Strata in the post-SR survey Stratum

Merged?

1 (non merged) 2 (Small – small part)

No Yes

3 (Small – large part)

Yes

4 (Large – small part) 5 (Large – large part) Total

Yes Yes

PreSR size N/A Below median Below median Above median Above median

Population growth

No. of municipalities

Share of sample

N/A Above median Below median Above median Below median

8 6

25,0 %

Share complete answers 25,1 % 18,7 %

18,7 %

6

18,7 % 18,8 %

6

18,7 % 18,8 %

6

18,8 % 18,8 %

32

100,0 %

100,0 %

38

Table 3: Questions, definitions, and descriptive statistics Variable Independent variables Vocational training College degree Master’s degree or above Income Publicly employed Gender Age Dependent variable: Internal political efficacy Qualified Understand Public office Informed Complex

Question or definition Respondent has completed vocational training Respondent has completed two year theoretical education Respondent has received Master’s degree or higher Pre-tax income of household in categories Employed in the public sector Gender of respondent (1= female) Age of respondent at time of survey Additive index of the following five variables I consider myself to be well qualified -to participate in municipal politics I feel that I have a pretty good under-standing of the important political issues facing our municipality I feel that I could do as good a job as a local councilor as most other people How well would you say that you are informed about local politics in your municipality? Sometimes local politics seem so complicated that a person like me can't really understand what's going on

N 2763 2763 2763 2281 2763 2763 2761 2450

Pre-SR survey Mean Std. dev. .25 .43 .19 .39 .08 .27 2.7 1.2 .36 .48 .52 .50 48 16 16 4.6

Min 0 0 0 1 0 0 17 5

Max 1 1 1 5 1 1 96 25

N 1233 1233 1233 1233 1233 1233 1233 1174

Mean .34 .26 .13 2.7 .25 .50 53 16

Post-SR survey Std. dev. Min .48 0 .44 0 .34 0 1.3 1 .43 0 .50 0 15 24 3.9 5

2716

2.8

1.6

1

5

1225

2.8

1.4

1

5

2612

3.8

1.2

1

5

1212

3.8

1.1

1

5

2660

2.9

1.6

1

5

1208

2.9

1.4

1

5

2728

3.4

.93

1

5

1233

3.2

1.0

1

5

2679

3.4

1.5

1

5

1213

3.2

1.3

1

5

39

Max 1 1 1 5 1 1 95 25

Table 4: Regression‐based DiD‐analysis of the effect of municipal amalmagation and size on political efficacy Dependent variable: Internal political efficacy Postreform (P = 1) ‐0.555 ‐0.557 ‐0.601 [0.258]** [0.258]** [0.257]** Treated municipalities (M = 1) 0.093 [0.292] Treated Ms, post‐reform (P x M = 1) ‐0.772 [0.281]** Treated municipalities (M = SS) 0.154 [0.402] Treated municipalities (M = SL) ‐0.089 [0.367] Treated municipalities (M = LS) 0.215 [0.435] Treated municipalities (M = LL) 0.095 [0.366] Treated Ms, post‐reform (P x SS = 1) ‐1.099 [0.339]*** Treated Ms, post‐reform (P x SL = 1) ‐0.545 [0.314]* Treated Ms, post‐reform (P x LS = 1) ‐0.888 [0.419]** Treated Ms, post‐reform (P x LL = 1) ‐0.573 [0.423] Population change from reform (in 10,000) ‐0.01 [0.058] Population change x P = 1 ‐0.158 [0.065]** Vocational training 0.006 0.009 ‐0.068 [0.213] [0.213] [0.196] College degree 0.72 0.714 0.598 [0.238]*** [0.238]*** [0.226]** Master's degree or above 2.161 2.153 2.07 [0.366]*** [0.370]*** [0.361]*** Income, 2nd quintile 0.652 0.655 0.487 [0.269]** [0.266]** [0.247]* Income, 3rd quintile 0.861 0.858 0.696 [0.311]*** [0.312]** [0.299]** Income, 4th quintile 2.331 2.322 2.157 [0.371]*** [0.369]*** [0.348]*** Income, 5th quintile 2.911 2.908 2.89 [0.475]*** [0.472]*** [0.424]*** Publicly employed 0.914 0.908 0.972 [0.223]*** [0.224]*** [0.198]*** Female ‐0.973 ‐0.975 ‐1.054 [0.167]*** [0.171]*** [0.154]*** Age  0.124 0.123 0.147 [0.034]*** [0.033]*** [0.035]*** Age^2 ‐0.001 ‐0.001 ‐0.001 [0.000]** [0.000]** [0.000]*** Observations 1992 1992 2251 R‐squared 0.15 0.15 0.15 Sample Full Full Full sample sample sample Robust standard errors corrected for clustering at the (new) municipal level in brackets. * significant at 10%; ** significant at 5%; *** significant at 1%.

40

Table 5: Matching DiD‐analysis of the effect of municipal amalmagation and size on political efficacy Exact treatment group matching on: Categorical treatment group Unweighted sample (n  = 1417) Weighted sample (n  = 3012) Binary treatment Categorical treatment Binary treatment Categorical treatment ATT ‐0.967 ATT_SS ‐1.879 ATT ‐1.273 ATT_SS ‐1.953 [0.411] ** [0.586] *** [0.425] *** [0.549] *** ATT_SL ‐0.695 ATT_SL ‐0.506 [0.445] [0.647] ATT_LS ‐1.150 ATT_LS ‐2.079 [0.525] ** [0.713] *** ATT_LL ‐0.133 ATT_LL ‐0.515 [0.752] [0.720] Exact treatment group matching on: Pre‐reform municipality Weighted sample (n  = 2794) Unweighted sample (n  = 1226) Binary treatment Categorical treatment Binary treatment Categorical treatment ATT ‐1.114 ATT_SS ‐1.751 ATT ‐1.608 ATT_SS ‐2.305 [.320] *** [.481] *** [.356] *** [.532] *** ATT_SL ‐0.838 ATT_SL ‐1.480 [.624] [.532] *** ATT_LS ‐1.077 ATT_LS ‐1.480 [.684] [1.267] ATT_LL ‐0.661 ATT_LL ‐1.048 [.376] * [.370] *** Notes: Matching‐procedure described in text. Standard errors, reported in parentheses, are corrected for clustering at the new    municipal level, bootstrapped with 50 replications and resampling at the cluster level.

41

Table 6: Retrospective evaluations of difference in political efficacy before and after reform. Binary treatment ‐1.376 ‐1.247 ‐1.291 0.313 *** 0.381 *** 0.378 *** Categorical treatment SS ‐2.099 ‐2.065 ‐2.158 0.455 *** 0.498 *** 0.613 *** SL ‐2.243 ‐2.022 ‐1.832 0.523 *** 0.489 *** 0.513 *** LS ‐1.223 ‐0.863 ‐0.827 0.301 *** 0.489 * 0.549 LL ‐0.472 ‐0.266 ‐0.394 0.272 * 0.559 0.546 Test for coefficient equality§ p = .000 p = .023 p = .063 Population difference, absolute ‐0.301 ‐0.29 ‐0.328   (measured in 10,000s) 0.054 *** 0.069 *** 0.077 *** Population difference, relative ‐0.184 ‐0.197 ‐0.25 0.061 *** 0.058 *** 0.065 Matching, binary treatment ‐1.555 ‐1.256 ‐1.307 0.348 *** 0.383 0.379 Bias corrected c.i. [‐.646;‐2.171] [‐.136;‐1.854] [‐.473;‐1.831] Full Unweighted Weighted  Sample sample matched matched sample sample No. of observations 1031 531 1352 § Reports p‐values associated with the test of the null hypothesis that the coefficients   of the treatment categories are equal against the alternative that they are not equal. Note: All standard errors are bootstrapped using 200 replications, accounting for             clustering at the post‐reform municipal level.

42

Figure 1: Distribution of propensity scores across treatment status and time Control, post-reform

Treatment, pre-reform

Treatment, post-reform

0 1 .5 0

Pr(treat)

.5

1

Control, pre-reform

0

.5

1

0

.5

1

Pr(postreform)

43

Table A: Municipalities in the sample

Stratum

Selected

Size

Size

Pop. Growth

municipalities

2006

2007

(percent)

501,158

503,699

0.5

Not merged

København

Not merged

Dragør*

13,154

13,184

0.2

Not merged

Gladsaxe

61,735

61,945

0.3

Not merged

Høje Taastrup

46,257

46,683

0.9

Not merged

Vallensbæk*

12,230

12,145

-0.7

Not merged

Helsingør

61,340

61,012

-0.5

Not merged

Ringsted

31,094

31,468

1.2

Not merged

Læsø

2,091

2,058

-1.6

Small before, now small part

Ramsø*

9,412

81,017

760.8

Small before, now small part

Holmegaard*

7,643

80,133

948.4

Small before, now small part

Glamsbjerg

5,924

41,816

605.9

Small before, now small part

Augustenborg

6,525

76,825

1077.4

Small before, now small part

Gram

4,867

56,275

1056.3

Small before, now small part

Lundtoft

6,150

60,044

876.3

Small before, now large part

Ledøje-Smørum

10,797

40,057

271.0

Small before, now large part

Jægerspris

9,520

43,910

361.2

Small before, now large part

Stenlille

5,634

28,956

414.0

Small before, now large part

Trundholm

11,311

32,980

191.6

Small before, now large part

Ejby

10,192

36,771

260.8

Small before, now large part

Hadsten

11,969

45,037

276.3

Large before, now small part

Frederikssund

19,144

43,910

129.4

Large before, now small part

Korsør*

20,873

76,949

268.7

Large before, now small part

Hedensted

17,190

44,892

161.2

Large before, now small part

Grenaa

18,673

38,333

105.3

Large before, now small part

Viborg

44,505

91,405

105.4

Large before, now small part

Hjørring

35,296

67,118

90.2

Large before, now large part

Hillerød

38,102

46,354

21.7

Large before, now large part

Middelfart*

20,599

36,771

78.5

Large before, now large part

Randers

62,524

92,984

48.7

Large before, now large part

Silkeborg

55,906

86,540

54.8

Large before, now large part

Skive

27,972

48,344

72.8

Large before, now large part

Aalborg

163,952

194,149

18.4

44